OpenAI Says Its AI Made 3 Scientific Discoveries, Plus a Cheaper Model

Ready to transform artificial intelligence from a buzzword into your personal revenue generator?

HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.

Inside you'll discover:

Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.

Get The Free Guide

1  OpenAI claims three scientific discoveries and ships a cheaper model

OpenAI's president presented three findings from an internal model, in math (harmonic analysis), computer science (complexity theory), and physics (quantum field theory). None has been independently verified yet. Earlier the same day, OpenAI released GPT-6.1 Terra, a faster and cheaper alternative to GPT-6 Astra that is especially good at using a computer. It's available on all paid plans in Work and Codex.

2  Microsoft offers buyouts to about 60,000 employees

Microsoft is reportedly offering voluntary buyouts to about a quarter of its workforce, with payouts based on tenure, continued stock vesting, and temporary healthcare. Staff reportedly have until early December to decide. The move follows Microsoft's disclosure of $116B in fiscal 2026 spending, most of it on data centers and equipment.

3  Anthropic shows how it reviews code when AI writes most of it

Anthropic says Claude now writes about 90% of Claude Code's code, and daily merges jumped from 2,500 in May to more than 30,000 in September. Engineer Addy Osmani and two colleagues explain how the team adapted. The biggest change: AI handles most reviews, so engineers can focus on the riskiest changes.

4  Researcher publishes 11 unpatched Tailscale flaws

Security researcher Erica Windisch published 11 flaws in Tailscale and the self-hosted headscale server. In her lab, one compromised device used SSH port forwarding to take over other machines and add a rogue device. Tailscale says fixes are coming, but there's no patch yet. Until then, review who has SSH access and turn off forwarding you don't need.

The filing. Ahead of its IPO, Anthropic published its S-1 filing, the first hard numbers the company has ever shared. Shares are expected to start trading the week of November 16.

Revenue is soaring. Anthropic brought in $16B in the first half of 2026, up 1,300% from a year earlier.

Losses are shrinking. Its operating loss was $693M, down from $2B a year before. The headline net loss of $81B comes mostly from a $72B accounting adjustment. Still, about $800B in future compute agreements are coming due.

Very lean. Anthropic earned $6.6M in revenue per employee over the past 12 months. SemiAnalysis calls it one of the most efficient scale-ups in history.

The risks. Nearly 100 pages cover what could go wrong, from authoritarian governments misusing its models to AI getting out of control.

Why it matters. How Wall Street reacts could shape funding for the whole industry. For a quick visual summary, see these six charts

Start from the deadline. GitHub staff engineer Sean Goedecke works backward from the timeline instead of forward from the task. Estimates decide which projects get funded, so the real question is what version of the work fits the time available.

His four steps. First, gauge how much pressure is behind the request. Second, start from the timeline and ask what fits. Third, add buffer for the parts of the codebase you don't know. Fourth, offer options with different levels of risk.

Protect your credibility. He saves "this is impossible" for rare cases, so people take it seriously when he says it. He also argues that engineers should be willing to commit to a position.

The catch. Estimates don't carry over well between teams. Some teams work 10 times faster than others, so a number that fits one team won't fit another.

📸 Deleted, Not Gone. One user says Claude Code found thousands of "deleted" photos still sitting in a hidden iPhone cache (2.5M views). Another found about 152,000 files the same way (14K bookmarks). Their fix: check which apps have Files and Folders access and remove the ones you don't need.

🦊 Throwback Ad. Firefox's latest ad uses early-internet nostalgia to pitch a more private browser where AI is optional (4.6M views).

₿ Steal My Bitcoin. A Hacker House co-founder put Bitcoin on a deliberately vulnerable PHP site to test whether the PHP panic is overblown (2.7M views). It was cracked within 24 hours, but the exploit took serious skill, including precise memory corruption and getting around memory protections.

🦠 Claude to the Rescue. Jeremy Cioara says Claude Code found malware on his Mac that was sending files to an attacker, then stopped it within about 15 minutes (4.4K likes).

🔓 Don't Lock It Down. Eric Raymond argues that banning public access to frontier models would fail, and that defenders should use the same tools to fix bugs first (4.8K likes).

🧪 A Language for Agents. Matt Pocock says today's languages weren't built for AI agents and points to Bend as an interesting start, though not ready yet (4.7K likes).

🏃 Run, Drone, Run. Purdue researchers built a drone that follows runners through forests without GPS (2.4M views). Skeptics point out that lost hikers don't usually run.

📚 AI Words to Know. CNET put together 65 AI terms in plain English.